Optical Reservoir Computing

16-node optical neural network. Credit: Ghent University

Reservoir Computing is a paradigm that mimics, to a certain extent, the way our brain processes and store information and learning in the process.

It is likely a paradigm for the "conoscenti" only but nevertheless one that can lead to interesting applications in the future to create emerging information.

In our brain we have communications among neurones that triggers change of state (out of a continuum, not in a binary sense where a change of state means flipping from 0 to 1 or vv, but rather making the receiving neurone more or less likely to fire...). Sometimes this state change (of one or several neurones) lead to a signal going to a motor nerve that activate a muscle and therefore an action. Other times (most of the times actually) the change of state in several neurone originate a conscious perception (a thought). Indeed, a though, something that we can consider an emerging information, is the result of a state change. And once you have had a thought that one seems to generate other thoughts and over time having had a thought will make you interpret incoming signals in a new way, that is you have learnt.

Reservoir Computing is a paradigm that leads to very similar result, although usually on a much smaller scale, because of physical infrastructure limitation. 
Researchers have invented various physical architectures to implement this paradigm. One example is that of neural networks.

Now a team of researchers at the Ghent University in Belgium are reporting on a paper on Nature Communications to have developed a novel physical infrastructure based on optical interconnections and photonics chips that implement this paradigm.

The advantage of this implementation is the increased speed, up to the hundreds of Gbps, and the very low power consumption.

It is interesting to see how technology provides new tools to mimic Nature and how scientists are finding new ways to flip the von Neumann paradigm into new processing architectures.

The reservoir computing is not going to displace von Neumann computation. Whem you ask "how much is it 2 plus 2, you rather expect a 4 as an answer, than any other one. However this is not what you get from a brain. Try asking hundred times your friend how much is 2 plus two and before reaching the hundredth time you will rather get as an answer not "4" but "Hey! What's going on? Are you joking? I already told you...". This is because your friend remember the past and that influence his reply.

However, there are many areas where getting different answers to the same input makes sense, it is what we call learning by experience, and that includes image and voice recognition, to mention but two of them.

Author - Roberto Saracco

© 2010-2018 EIT Digital IVZW. All rights reserved. Legal notice